The Moravec Paradox is Not Paradoxical
Intelligence from the Primordial Soup of Signs
Encoded in the large, highly evolved sensory and motor portions of the human brain is a billion years of experience about the nature of the world and how to survive in it. The deliberate process we call reasoning is, I believe, the thinnest veneer of human thought, effective only because it is supported by this much older and much more powerful, though usually unconscious, sensorimotor knowledge. We are all prodigious olympians in perceptual and motor areas, so good that we make the difficult look easy. Abstract thought, though, is a new trick, perhaps less than 100 thousand years old. We have not yet mastered it. It is not all that intrinsically difficult; it just seems so when we do it.
Hans Moravec, Mind Children
The “Moravec Paradox” is not a paradox at all: it is the result of bad philosophizing. In other words, unexamined assumption, unconscious metaphor, and self-congratulatory conclusions have plagued this field just as they have many other fields which depend upon shallow psychologizing and naïve introspection.
Intelligence does not reach its apex and most essential expression in abstract reasoning: a truth table is not the heart of intelligent behavior. Proprioception, creative manipulation, and navigation in a complexly textured real world are the essential challenges of intelligence and therefore the foundation of all other forms of intelligence, including symbolic reasoning.
But how could it be, that symbolic capacity originates in perceptual and sensorimotor tasks? How could being-in-the-world result in reasoning-about-the-world? What’s the probable evolutionary genealogy?
Let’s consult ethology for a moment and consider the example of the “supernormal sign stimulus”. As Nikolaas Tinbergen famously demonstrated, an animal is not only fooled by a substitute sign, but will respond proportionally to a sign’s exaggerated character: the baby seagull will respond all the more vigorously to the minimal but exaggerated sign stimulus of an unnaturally bright red dot on the beak of its feeding parent.
What can this tell us? That the activating gestalt for an instinctual behavior can be isolated and toyed with, because instinctual discharge is already inherently symbolic. As the immensely influential but still largely unknown Jakob von Uexküll convincingly showed, every animal lives within its own Merkwelt: there is no “raw” data, but a curved field of vectors and weighted signs.
The animal’s own world is not only dependent on what its sense organs can or cannot receive. Its sensory world is still more restricted; it is composed of sign stimuli […]
N. Tinbergen, Study of Instinct, p.37
These ideas form the basis of the field known as “biosemiotics” - a name encompassing several scholarly threads all simultaneously promising, disappointing, and neglected. But what it can offer to the hype-drenched field known as AI, is the realization that the “sensorium” is neither a tabula rasa as per the vestigial Cartesian-Kantian psychology which still dominates the cognitive sciences, nor solely composed of “always-already” signs as per the fatalist French poststructuralism which dominates the academic humanities, but the undulating inner surface of a sphere of potential signs. The world of the most-living is itself squirming with life: not “always-already” but “maybe-now”. Anyone remotely familiar with peak cortical performance - or just serotonin overdose as induced by psilocybin - should know what I’m talking about: the inner surface of the Merkwelt is not merely “data-driven”, it is the seething primordial ooze of meaning. Generating and sustaining emergent coherence from this sensory soup, such that a bottomup optimization escapes the obtuse blindness which every topdown approach succumbs to, is what should be properly considered the “hard problems” of intelligence.
But we should not assume that such a description of the organic fails the test of rigor: did you know that proteins vibrate according to their macrostructure? The processes which animate the cytoplasm should not be thought of as though chemistry proceeded like a Turing tape: it’s more like a Gamelan orchestra, a throbbing cadence, a vast unthinkable symphony of ratio. After all, the ratios which characterize chemical reactions are due to the basins of attraction which establish the orbital nodes of electron clouds, and thus the same interval structure - otherwise known as the “harmonic” - percolates upward into the great cacophony of the organic, producing yet more emergent structure built from constraint and dissipative flow. Such is how I experience the senses: neuromorphic experience is not organized by intelligence, it generates intelligence from its own dalliances with chaos.
Sense data is a heavily weighted graph, with strong attractors and repulsors: to “sense” is to traverse a field of virtual signs which actuate in degrees. Every creature is both constantly hallucinating and overreacting, and constantly ignoring and remaining passive: the question is, what matters to this creature, at this time, in this place.
To build anything like a viable hypothesis of intelligent behavior requires that we internalize this lesson and allow the symbolic reasoning to percolate from the lowest possible levels: the more a symbol is defined in terms of recent and accurate sensorimotor feedback, the more meaningful it is. In other words, abstraction is obtained with a proportional sacrifice of relevance: most of the apparently symbolic behavior of a living creature is only visible due to frustration and displacement. What is more deeply symbolic looks like “purpose”, because it’s hidden behind success. When a bird builds a nest by gathering twigs, we say it does this because it’s nesting time. When it tosses twigs around in a courtship display, we say it’s doing something symbolic and imply uselessness: but the real symbolic moment is rooted much earlier and deeper within the avian adaptive package...
It may aid us to realize that most of human culture consists of the trade in “supernormal sign stimuli”. The pornography of instinct: hunting, killing, eating, mating, nesting, brooding, and so on are all exploited and given supernormal signs in cultural activity. A particular Dogon mask evokes a curious reaction of fear, loathing, and desire: why? Because it activates several sign systems at once with a supernormal intensity. Arguably, much of the human art of psychologizing, is the supernormal stimulus of social signaling for the purpose of acquiring power over the target: psychology is a weapon, and the urform of the “syndrome” is the mask. A mask calls a spirit: and a spirit is, from this perspective, nothing but the coagulation of excess social response into a relatable form. Tribal apes deal with “persons”, therefore anything which must be dealt with, must first become a persona: thus each spirit has its mask, and each mask has its power of supernormal stimulus. We are, as the most teachable techno-ape, not only “freed” from instinct, and thus prone to instinctual incoherence and thus anxiety and thus neurosis, but gifted and cursed therefore with an excess of instinctual energy without unambiguous aim: we become both more excitable via supernormal stimulus because our instincts are weakened in aim, and numb to stimulation because we do it to each other almost incessantly. In postagricultural civilization the acceleration becomes unsustainable. Every postagricultural culture is an orgy of mounting overstimulation and an arms race between the subtle arts invoking a set of advantageous responses, and the exhaustion of those responses: the pyramids grew in size until they no longer inspired the Egyptians to be Egyptian. Greek austerity eventually was not a reprieve from Greek sensuality, but merely boring, because they were no longer vivacious enough to crave the contrast.
They’ve been going about it backwards - these artisans of artificial intelligence. Symbolic reasoning is not the prerequisite to intelligent behavior, it is the consequence of coherent sensorimotor preprocessing.
Every object choice is a symbolic object choice, only made possible by rigorously refined and amplified data propagation. A “symbol” is what remains of a signal after iterative filtering. Therefore stochastic resonance is likely at play, everywhere perception is at work: an optimum noise complement makes relevance discoverable, because otherwise a sensory field could not anticipate vital signals fast enough. Thus it is that sensory deprivation will yield up the baseline hallucination which forms the bedrock of all perception: this is how dreaming was born, and why it became yet another opportunity to refine the same processes of sensory-symbolic generation… Freud was absolutely correct that the dreamstate is characterized by a tightly packed vector field of symbolic valence, but wrong that its traversal is motivated by mere “wish”: the organism wastes no opportunity to rehearse its primary challenge - which is to navigate its own world of signs, to swim confidently through its niche for a lifetime, exploring every cranny of its ancestral mind.
Likewise, we only know what anything “means” because of the almost incorruptible coherence of the lower levels of sensory processing: we are presented with meaning, and then ascribe to the process a mysterious agency like “intention”. But almost all conscious reasoning is ex post facto reasoning: we fabricate memorable tales to tell each other, and only as an afterthought do we spin the yarn to ourselves. The “I” is a consequence of sociality and was invented long after the You and the We. The individual appears only after the tribe is millennia old.
More from this series: